OpenAI workspace agents
- OpenAI added “workspace agents” to ChatGPT for Business, Enterprise, Edu and Teachers so teams can build custom bots that run workflows. - These agents can perform background tasks in the cloud and replace Custom GPTs for team use, powered by Codex. - The feature packages agents as a team productivity layer that surfaces questions about safe, agent-driven workflows across systems (theverge.com).
OpenAI has added shared “workspace agents” to ChatGPT, letting teams build bots that can run work across company tools in the background. (openai.com) OpenAI said the feature launched April 22, 2026 for ChatGPT Business, Enterprise, Edu, and Teachers. The company said the agents are powered by Codex, can be shared inside a workspace, and can keep running in the cloud after a user closes ChatGPT. (openai.com) The new agents are meant for repeatable jobs, not one-off chats. OpenAI said teams can connect them to Slack, Google Drive, Google Calendar, Microsoft SharePoint, and custom Model Context Protocol servers, then schedule runs or use them inside ChatGPT and Slack. (openai.com) In plain terms, OpenAI is turning ChatGPT from a question box into a workflow tool. Instead of answering a prompt and stopping, these agents can pull files, follow steps, hand work across apps, and return results later. (openai.com) That shift also changes what “custom bots” mean inside ChatGPT. OpenAI said workspace agents are “an evolution of GPTs,” and its help documentation says they replace Custom GPTs for team workflow use in Business and Enterprise workspaces. (openai.com 1) (openai.com 2) The sales pitch is productivity with guardrails. OpenAI said workspace agents operate within organization permissions, while admins can control enabled apps, role-based access, and connector settings across Business, Enterprise, and Edu plans. (openai.com 1) (openai.com 2) OpenAI is tying the product closely to Codex, its agent system for software and tool use. On OpenAI’s developer site, Codex is described as an agent that works across tools and cloud environments, with customization layers such as persistent instructions, reusable skills, subagents, and external-tool access through Model Context Protocol. (openai.com 1) (openai.com 2) That makes the release part software feature, part governance test. Once an agent can read shared drives, send messages, and run on a schedule, companies have to decide which systems it can touch, which roles can use it, and how much autonomy to allow. (openai.com) (openai.com) OpenAI said the rollout will happen over the next few weeks for Business and Enterprise workspaces, with sales contact for broader deployment. The product opens with a simple promise: build one agent for the team, then let it keep working after the chat ends. (openai.com)
